The Time Course of Neural Activity Predictive of Impending Movement.

We consider the problem of uncovering the time course of neural activity leading to self-initiated movements, and, relatedly, of the degree to which such activity can be used to predict when a movement is imminent. We address a major pitfall common across decades of experimental research on self-initiated movement, namely that the data epochs subjected to analysis are only those culminating in a movement. We propose a rigorously controlled experimental paradigm that yields epochs that either do or do not terminate with a self-initiated movement but are well-matched in other respects. We applied this framework as a basis for experiments in which we recorded M/EEG data from human participants, and then used machine learning in a sliding window to classify data segments as belonging to a movement or non-movement epoch. When we tried to classify a window of data as belonging to a remote baseline versus some other temporal offset closer to movement, as is currently standard in the field, the results suggested movement could be accurately predicted even 1.5 seconds before movement onset. By contrast, when we instead included no-movement epochs, this kind of extreme result was abolished; rather, we observed that classification accuracy rose abruptly from near chance to near ceiling as movement onset approached, and at time scales more in line with what participants report in such tasks. As these and further analyses show, our framework provides an improved vantage point for studying fundamental characteristics of neural activity specifically directed toward self-initiated movement.